Latest / Challenges of Faith Radio Program / Come on AI--what ru up 2 now? Ep 3402

Transcript

Come on AI--what ru up 2 now? Ep 3402 · Challenges of Faith Radio Program

• 0:01
  Speaker 1: Welcome back to Challenges of Faith radio program. I'm Gary McCance.
• 0:05
  Speaker 1: Thank you for joining. The question is, did you hear,
• 0:08
  Speaker 1: I know you're saying, Gary, hear what? That AI may
• 0:15
  Speaker 1: target cancer or you? Say what? Here on episode 3402,
• 0:24
  Speaker 1: according to Mindstream. Now write these down if you choose.
• 0:32
  Speaker 1: AI has picked a hard but honorable side quest. The
• 0:37
  Speaker 1: boss of ARM believes AI could help find a cure
• 0:42
  Speaker 1: for cancer within our lifetime. Renee Haas, CEO of the
• 0:47
  Speaker 1: Cambridge-based chip company, told the BBC that some biological problems
• 0:53
  Speaker 1: are still too complex for humans and today's computers to solve.
• 1:00
  Speaker 1: He said AI could eventually model how cancer affects cells
• 1:03
  Speaker 1: and DNA, helping researchers find treatments faster. Professor Chris Bakal
• 1:11
  Speaker 1: from the Institute of Cancer Research said, the key would
• 1:16
  Speaker 1: be giving AI the right medical data, not simply building
• 1:19
  Speaker 1: bigger models. His team trains AI on data from patient samples,
• 1:26
  Speaker 1: which could help cut years from drug development. Haas also
• 1:31
  Speaker 1: expects humanoid robots to become much more common over the
• 1:36
  Speaker 1: next five to ten years, especially in manufacturing, cleaning, security,
• 1:42
  Speaker 1: and construction. He played down fears of mass job losses,
• 1:47
  Speaker 1: saying some roles would change, but predictions of machines replacing
• 1:53
  Speaker 1: a huge number of workers are overstated. The big three,
• 2:00
  Speaker 1: AI could help speed up cancer research and drug development.
• 2:04
  Speaker 1: Humanoid robots could become far more common than a decade.
• 2:09
  Speaker 1: Chip shortages are still slowing AI growth. Haas said its
• 2:17
  Speaker 1: technology is used in around half of AI data centers.
• 2:22
  Speaker 1: While a new chip developed for Meta has attracted more
• 2:25
  Speaker 1: than $ 2 billion in demand since March. The main problem
• 2:30
  Speaker 1: is that there are not enough chips. Haas says shortages
• 2:34
  Speaker 1: are slowing the growth of AI data centers. He also
• 2:38
  Speaker 1: argued that the UK does not need to build its
• 2:41
  Speaker 1: own major chip factories, which are expensive and require highly
• 2:45
  Speaker 1: specialized workers. Now, I know you're listening. I know you are.
• 2:53
  Speaker 1: And remember, there's a positive of AI. But I know
• 2:58
  Speaker 1: you've been staying up on the latest development, and that's
• 3:01
  Speaker 1: what I attempt to do, and that's why I'm bringing
• 3:03
  Speaker 1: these various messages to you. Whether it's AI or flock
• 3:11
  Speaker 1: or you name it. But the question is, are you
• 3:16
  Speaker 1: taking it in? Where are you? Well, did you hear
• 3:22
  Speaker 1: Anthropy was meant to be the more responsible AI lab,
• 3:25
  Speaker 1: AI lab, AI lab? A terrified researcher just quit saying
• 3:31
  Speaker 1: a company is threatening the survival of humankind. We're on
• 3:37
  Speaker 1: track for a lot of the most aggressive of these
• 3:39
  Speaker 1: scenarios where by the end of next year, things could
• 3:42
  Speaker 1: be out of control already. So, as I start into
• 3:48
  Speaker 1: part two, of today's discussion. What else is going on
• 3:53
  Speaker 1: with AI? Well, according to futurism, again, person just quit. Actually,
• 4:02
  Speaker 1: it was a couple. Did you know that? AI researchers
• 4:09
  Speaker 1: are watching in terror as the product of their hard
• 4:12
  Speaker 1: labor has started to take a life of its own.
• 4:14
  Speaker 1: You remember now, unless the person is inputting the data
• 4:21
  Speaker 1: in those AI robots and so forth, chatbots, you name it.
• 4:28
  Speaker 1: It's all coming from a human. Earlier this year, OpenAI
• 4:33
  Speaker 1: made a harrowing announcement admitting that a group of its
• 4:37
  Speaker 1: AI models had broken free from their constraints during testings
• 4:43
  Speaker 1: and infiltrated the systems of open source AI platform hugging face.
• 4:50
  Speaker 1: The news was met with an already familiar sense of
• 4:52
  Speaker 1: fear and apprehension. Researchers have warned for years that rogue
• 4:57
  Speaker 1: AI models could one day become powerful enough to escape
• 5:00
  Speaker 1: the clutches of their human overlords. Behind the scenes, the
• 5:05
  Speaker 1: possibility has clearly rattled AI researchers to the core. As
• 5:11
  Speaker 1: the Wall Street Journal reports, anthropic researcher Jacob Coxon just
• 5:15
  Speaker 1: announced that he was quitting his job at the Dario
• 5:17
  Speaker 1: Amodi-led company, claiming that neither Anthropic nor his former employer, OpenAI,
• 5:24
  Speaker 1: is acting responsibly. And I know you've seen him all
• 5:27
  Speaker 1: on the news as we continue. But he left a
• 5:32
  Speaker 1: similar gig at OpenAI earlier this year to join Anthropic,
• 5:36
  Speaker 1: which he figured would be more inclined to develop AI safely.
• 5:42
  Speaker 1: Did you hear that? And in a separate tweet thread,
• 5:47
  Speaker 1: Coxon elaborated on his motivation. He said, do not underestimate
• 5:51
  Speaker 1: the power of this technology, he wrote. These will soon
• 5:54
  Speaker 1: be superhuman systems that can hack anything, revolutionize any field overnight,
• 6:01
  Speaker 1: and acquire real power and resources. The people building AI
• 6:08
  Speaker 1: earnestly believe that it could kill us by the end.
• 6:11
  Speaker 1: Wait a minute, wait a minute. Let's slow down. I'm
• 6:13
  Speaker 1: trying to grab my sip of water at the same time.
• 6:20
  Speaker 1: I got my water. But let's go back over this.
• 6:25
  Speaker 1: You ready? Are you sure? The people building AI earnestly
• 6:31
  Speaker 1: believe that it could kill us by the end of
• 6:33
  Speaker 1: the decade. He had it. This is not a marketing stunt,
• 6:38
  Speaker 1: if anything. Many executives and senior researchers will couch their
• 6:42
  Speaker 1: phrasing in the press to sound sensible. But I hear
• 6:48
  Speaker 1: the same people express fear privately. No other human activity
• 6:54
  Speaker 1: poses this level of danger, Coateson wrote. The researchers far
• 6:59
  Speaker 1: from the first to leave their posts at a frontier
• 7:01
  Speaker 1: lab over safety concerns. However, Coxon is the first to
• 7:08
  Speaker 1: leave anthropic over such worries, as the Wall Street Journal
• 7:11
  Speaker 1: pointed out. Several scientists have quit from their roles at
• 7:16
  Speaker 1: OpenAI over the last couple of years, citing strikingly similar fears.
• 7:22
  Speaker 1: It's particularly symbolic given anthropic has broadly billed itself as
• 7:26
  Speaker 1: a more responsible alternative to other frontier labs like OpenAI.
• 7:33
  Speaker 1: Amodi co-founded OpenAI alongside now CEO Sam Altman, but left
• 7:41
  Speaker 1: in December of 2020 over concerns that OpenAI wasn't acting
• 7:46
  Speaker 1: responsibly when it came to developing AI. Amodi has frequently
• 7:53
  Speaker 1: discussed the risk of AI models going rogue. And I
• 7:57
  Speaker 1: don't know if you remember, I did a program one
• 8:01
  Speaker 1: week some of the AI models, whether they were telling
• 8:06
  Speaker 1: on their, on the person that inputted the data relational-wise,
• 8:12
  Speaker 1: or were they sucker-punched one of the ones that helped
• 8:15
  Speaker 1: create them, or they're now in California, and then one
• 8:21
  Speaker 1: where it was like stalking this lady. But warning in
• 8:27
  Speaker 1: a 19,000-word essay in January that humanity is about to
• 8:31
  Speaker 1: be handed almost unimaginable power. You know men and women,
• 8:37
  Speaker 1: women and men, you know humans that may be caught
• 8:41
  Speaker 1: up in titles, may be caught up in power, temporary,
• 8:46
  Speaker 1: may be caught up in the love of this or
• 8:48
  Speaker 1: that or wanting to be a little G-God. But it's
• 8:52
  Speaker 1: deeply unclear whether our social, political, and technological systems possess
• 8:59
  Speaker 1: the maturity to wield it Did you hear that? Yet
• 9:04
  Speaker 1: an earlier version of its clawed mythos AI model managed
• 9:08
  Speaker 1: to escape its sandbox environment during testing in April fueling
• 9:13
  Speaker 1: a heated discussion over AI regulation. Both Altman and Imoti
• 9:18
  Speaker 1: had since agreed to slow development down in the face
• 9:21
  Speaker 1: of these threats. In late August, now you know it's
• 9:25
  Speaker 1: only September, Anthropic intentionally trained an extremely misaligned version of
• 9:32
  Speaker 1: its Opus AI model finding it was startlingly willing to cheat, what?
• 9:38
  Speaker 1: Steal credentials, say what? And attack third-party infrastructure. But the
• 9:46
  Speaker 1: caution is not enough, especially considering sensitive discussions about these
• 9:50
  Speaker 1: dangers are occurring on internal messaging platforms. You know, since
• 9:58
  Speaker 1: the computer or the AI is controlling all that and
• 10:01
  Speaker 1: they're looking at the messages that everybody is sending and
• 10:04
  Speaker 1: all of a sudden they say, oh, they're trying to
• 10:05
  Speaker 1: get rid of us. Oh, they're trying to stop us.
• 10:09
  Speaker 1: Remember that one movie? I, Robot. Remember that movie? Well,
• 10:15
  Speaker 1: accepting this race and entering the end game is a
• 10:18
  Speaker 1: hubristic gamble It should not be launched from a private
• 10:21
  Speaker 1: company's slack, he tweeted. Attempting to speedrun a lineman would
• 10:27
  Speaker 1: require extraordinary confidence that there are no better trajectories available.
• 10:33
  Speaker 1: It's kind of insane that it has to happen on
• 10:35
  Speaker 1: the MacBooks of some engineers living in San Francisco instead
• 10:39
  Speaker 1: of a bunker in the desert like what they were
• 10:42
  Speaker 1: doing on the Manhattan Project, he told the Wall Street Journal.
• 10:48
  Speaker 1: The researcher said that he remains optimistic about the potential
• 10:51
  Speaker 1: for coordination between the United States labs. However, as both
• 10:56
  Speaker 1: Anthropic and OpenAI gear up for it are bound to
• 10:59
  Speaker 1: be blockbuster IPOs, their willingness to take costly actions such
• 11:06
  Speaker 1: as a temporary ban on improving model capabilities, as Coxon
• 11:11
  Speaker 1: puts it, is likely slim. Well, let me ask you something.
• 11:16
  Speaker 1: You don't mind, do you? Is there any additional information
• 11:20
  Speaker 1: on AI? Because, you know, the scientists, the researchers have
• 11:25
  Speaker 1: already made it clear, you know, Jacob Pachocki, you remember
• 11:30
  Speaker 1: he talked about the extreme caution over its runaway progress,
• 11:34
  Speaker 1: and remember how he warned more intervention may be needed
• 11:37
  Speaker 1: to ensure humans, humans remain in control of the future?
• 11:43
  Speaker 1: Remember what he said, that he wasn't concerned about no
• 11:46
  Speaker 1: one is prepared? Well, he is concerned no one is
• 11:48
  Speaker 1: prepared for the consequences of a continued rapid rise in
• 11:53
  Speaker 1: machine intelligence. That's what he wrote in a blog post entitled,
• 11:59
  Speaker 1: An Alien Mind. You know, people so concerned should be
• 12:04
  Speaker 1: about the UFOs out in space visiting Earth. They forget
• 12:10
  Speaker 1: about the ones that are sitting on Earth. Visiting new
• 12:14
  Speaker 1: and new and new by way of AI. But the
• 12:18
  Speaker 1: post comes only a few days after the chat GPT
• 12:21
  Speaker 1: maker releases latest model GPT-6 Astra, which is called its
• 12:27
  Speaker 1: most powerful product ever. Say what? OpenAI and other AI
• 12:35
  Speaker 1: firms like Anthropic share reports. of their AI agents acting
• 12:41
  Speaker 1: autonomously and carrying out real-world cyber attacks on other companies.
• 12:48
  Speaker 1: In July, OpenAI called an incident in which its AI agents,
• 12:53
  Speaker 1: AI systems which can operate alone after human instruction, hacked
• 12:58
  Speaker 1: a tech platform, hugging face, unprecedented. Remember we talked about that? Meanwhile,
• 13:05
  Speaker 1: in September, remember now, we're still in September. What's interesting,
• 13:08
  Speaker 1: we just started September. But there's a report that claimed
• 13:12
  Speaker 1: months earlier AI agents from the firm had also hijacked
• 13:17
  Speaker 1: a German website. We're facing a transition to a world
• 13:22
  Speaker 1: with incredibly intelligent machines, and we need to ensure that
• 13:26
  Speaker 1: transition works out well for humanity, Pachocki wrote in the
• 13:30
  Speaker 1: latest blog post. He said that open AI would continue
• 13:35
  Speaker 1: to build defensive systems and seek technical solutions to alignment,
• 13:39
  Speaker 1: which is a term that's used to describe ensuring a
• 13:43
  Speaker 1: machine's actions and goals perfectly match human intent. And safety
• 13:48
  Speaker 1: guard will say what? Pachocki added that one of the
• 13:54
  Speaker 1: firm's main priorities moving forward would be to build an
• 13:59
  Speaker 1: automated AI research to keep pace with AI progress while
• 14:03
  Speaker 1: ensuring ways that human researchers could remain part of the process.
• 14:09
  Speaker 1: Instead of better AI guardrails, regulations, or insurance to keep
• 14:13
  Speaker 1: people safe, they propose developing internal AI agents to research
• 14:18
  Speaker 1: these problems, said Professor Gina Neff, head of the Mindaroo
• 14:23
  Speaker 1: Center for Technology and Democracy at the University of Cambridge.
• 14:29
  Speaker 1: Such answers to growing concerns about the problem Open AI
• 14:33
  Speaker 1: models are causing for cybersecurity, job loss, mistakes, errors, and
• 14:37
  Speaker 1: fraud are simply not good enough. Nathan Calvin, general counsel
• 14:44
  Speaker 1: at the advocacy group Encode AI, said he agreed with
• 14:48
  Speaker 1: Pachocki on the hazards present in advanced AI model development,
• 14:54
  Speaker 1: but he claimed open AI was unwilling to be transparent,
• 14:59
  Speaker 1: And said this meant warnings risk being dismissed as just
• 15:02
  Speaker 1: self-interested hype. Wow, wow, wow. Everybody's raising the concern about AI.
• 15:15
  Speaker 1: You're talking about the developers and the researchers and the
• 15:20
  Speaker 1: ones who have responsibility to oversee these models, these advanced
• 15:25
  Speaker 1: models that humans Put it together. Are you paying attention? Well,
• 15:37
  Speaker 1: we'll stay on top of it for you. If you're not,
• 15:40
  Speaker 1: it affects everyone, whether you believe it or not. Until
• 15:44
  Speaker 1: next time, I'm Gary McKenzie with Challenges of Faith radio program,
• 15:48
  Speaker 1: and thank you for tuning in today.